Fast and quantitative noncontact laser ultrasound tapping detection of debonding in aerospace honeycomb sandwich panel based on autoencoder-softmax

Abstract

Aerospace grade honeycomb sandwich panels (HSPs) feature ultra-thin skins and honeycomb walls and thus are prone to debonding defects during manufacturing and service. A fast, non-destructive, and noncontact laser ultrasound tapping method combining the local fine C-scan imaging and the global sparse C-scan is proposed to detect the debonding in the ultrathin aerospace HSP. Firstly, by measuring the thermoelastic laser-induced vibration signals with fine C-scan at a small-scale region including both known intact and debonding defects, an automatic labelling algorithm is proposed to construct the dataset for training the Autoencoder (AE)-Softmax model. Then, based on the trained AE-Softmax model, the sparse C-scan only at the centroid of each honeycomb cell can quickly identify suspicious defects with low credibility in the HSP. Further, the suspicious cells in HSP are fine scanned to differentiate the intact or debonding status according to the area proportion of the connected component in the C-scan image. Finally, experiments are carried out in a second HSP to validate the proposed method, that all the four diversified defects, including multiple-debonding cells, one debonding wall, and adhesive removals, are successfully detected without false alarm, and the detection efficiency has been improved over 100 times compared with the conventional dense C-scan imaging.